Open quietlychris opened 2 years ago
This can be done using the above through the TCP connection, but in practice should be probably be addressed using UDP. The branch udp
is currently being developed in support of this, which should allow for the Msg::SET
to be used across networks with poor connectivity.
Do we currently have state transitions for Node<Tcp, Active, T>
back to Node<Tcp, Idle, T>
? That could allow us a fallback, and then maybe have a fallback in terms of connection policy for the Node attempting to reach back out to the Host.
Right now, if Nodes stay up, but the Host crashes, the Node will not attempt to re-connect, even if the Host becomes available again. This could be fixed through better state management; if the Node doesn't get a reply from the Host within a certain amount of time, it will reset to
Idle
, and begin periodically attempting to reconnect to the Host at the original address. There might be other ways of doing this as well; find some reading on how other middlewares do this sort of thing.